In a conventional dynamic random access memory (DRAM) system, a read training is required to be performed to optimize timing between data signals and a data strobe signal. When the read training operation is executed, however, all of the data pins (DQ pins) are required to be toggled or driven to obtain the timing information, causing more power consumption.